numr 0.6.0-beta.1

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

//! Shared utilities for reduction backward implementations

use crate::error::Result;
use crate::runtime::Runtime;
use crate::tensor::Tensor;

/// Ensure a tensor is contiguous, making a copy if necessary.
#[inline]
pub(super) fn ensure_contiguous<R: Runtime>(tensor: Tensor<R>) -> Result<Tensor<R>> {
    if tensor.is_contiguous() {
        Ok(tensor)
    } else {
        tensor.contiguous()
    }
}